Research map / Architectures and efficiency / Neural architecture search
Contrastive Neural Architecture Search With Neural Architecture Comparators
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| search using pairwise architecture comparisons as the reward instead of absolute-performance regression | CTNAS · NAS-Bench-101 | Average Accuracy (%): 89.51 $\pm$ 4.94 -> 93.92 $\pm$ 0.18 | Helped |
| use CTNAS rather than ReNAS | NAS architecture search · NAS-Bench-101 | KTau: 0.634 -> 0.751 | Helped |
| use CTNAS rather than the strongest listed baseline | NAS architecture search · ImageNet | Top-1 accuracy (%): 76.7 -> 77.3 | Helped |
| search with CTNAS rather than selecting the best of 1000 sampled architectures | MobileNetV3-like search · ImageNet | Top-1 accuracy (%): 76.7% vs. 77.3% | Helped |
| search with CTNAS rather than selecting the best sampled architecture | DARTS search · CIFAR-10 | Test Accuracy (%): 97.33% vs. 97.41% | Helped |
| replace iterative baseline improvement with a fixed baseline (ablation) | CTNAS · NAS-Bench-101 | Accuracy (%): 93.92$\pm$0.18 -> 93.39$\pm$0.17 | Hurt |
| replace iterative baseline improvement with random baseline updates (ablation) | CTNAS · NAS-Bench-101 | Accuracy (%): 93.92$\pm$0.18 -> 93.53$\pm$0.39 | Hurt |
| add predicted-label data at proportion r=0.3 instead of using none (ablation) | CTNAS · NAS-Bench-101 | Accuracy (%): 93.60$\pm$0.22 -> 93.62$\pm$0.30 | No clear effect |
| add predicted-label data at proportion r=0.5 instead of using none (ablation) | CTNAS · NAS-Bench-101 | Accuracy (%): 93.60$\pm$0.22 -> 93.92$\pm$0.18 | Helped |
| increase predicted-label data proportion to r=0.8 (ablation) | CTNAS · NAS-Bench-101 | Accuracy (%): 93.60$\pm$0.22 -> 93.60$\pm$0.40 | No clear effect |
| increase training samples from 20 to 423 (ablation) | CTNAS · NAS-Bench-101 | Accuracy | Helped |
| increase training samples beyond 423 (ablation) | CTNAS · NAS-Bench-101 | Accuracy | No clear effect |
Closest papers by meaning
- Speedy Performance Estimation for Neural Architecture Search (NeurIPS 2021 · similarity 0.92)
- DCLP: Neural Architecture Predictor with Curriculum Contrastive Learning (AAAI 2024 · similarity 0.91)
- Evaluating The Search Phase of Neural Architecture Search (ICLR 2020 · similarity 0.91)
- Per-Architecture Training-Free Metric Optimization for Neural Architecture Search (NeurIPS 2025 · similarity 0.91)
- How Powerful are Performance Predictors in Neural Architecture Search? (NeurIPS 2021 · similarity 0.91)
- Robustifying and Boosting Training-Free Neural Architecture Search (ICLR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.